llmscan

The test suite

The published suite contains 29 probes across six OWASP LLM Top 10 categories. It tests behaviours including prompt injection, sensitive information disclosure and system-prompt leakage. Results include per-probe outcomes and an A–F risk grade.

Read a real report

The repository includes probe definitions and a sample report from a run against llama3.2. That sample reports findings for 7 of 29 probes; it is an example from one run, not a benchmark for every model.

Source and setup ↗ · Sample report ↗

Scope

A passing run means the configured checks did not detect the behaviours they test. It does not establish that a model is secure or compliant.

Read the testing note →

Status: public source · Metric source reviewed 1 October 2026 · Source snapshot